【作者】 王义文；

【导师】 刘献礼；

【作者基本信息】 哈尔滨理工大学 ， 机械制造及其自动化， 2010， 博士

【摘要】 轴承是机械基础部件,而钢球作为滚动球轴承的关键零件,其表面缺陷情况直接影响轴承精度、动态性能和使用寿命。因此,对钢球表面缺陷检测技术的研究是具有非常重要的理论和实用价值。本文对基于图像技术的钢球表面缺陷检测仪开发中的光源选择、展开机构动力学仿真分析、钢球表面缺陷的模式识别等关键技术问题进行了深入的研究,并搭建了能够实际应用的样机,其主要研究内容如下:进行了检测系统的光源优化研究。从钢球表面反光特性的分析入手,研究了钢球表面成像的难点,建立了钢球表面的光反射模型。通过对大面积漫反射平板光源、漫反射扁平环形光源、漫反射球面光源和同轴光源等LED光源大量的理论和实验分析,最终确定了由FPR光源、LDR光源组合的检测机构照明方案。该光源方案,有效地解决了光晕现象、周围景物映入等问题,提高了钢球图像的质量和有效检测面积,为后期的图像处理奠定了基础。运用UG与ADAMS联合建立了检测系统展开机构的模型并进行了运动学和动力学仿真,较为真实的仿真出钢球在展开盘检测腔中的实际运动轨迹、受力及碰撞情况。钢球与检测腔侧壁存在碰撞导致钢球产生回弹运动,通过优化展开腔的直径、阻尼特性、摩擦盘与展开盘转速、摩擦盘搓动速度等结构参数和运动参数,可以改变钢球的运动状态,从而保证钢球表面能完全展开和检测效率最高。进行了钢球缺陷识别及分类关键技术研究。首先研究了钢球表面图像采集及降噪增强的图像预处理方法。将原始图片经过两次小波消噪处理消去高频白噪声,再经图像平滑处理使消噪后的图片平滑,然后设定灰度阈值运用Canny算子对图片进行边缘检测,最后对图片进行形态学处理以及图像锐化处理,为钢球缺陷的特征提取奠定了基础。确定将缺陷面积、缺陷长短径比、缺陷周长以及欧拉数等作为钢球表面缺陷识别的特征参数,并提出了一种基于BP神经网络的钢球表面缺陷类型识别方法。通过对采集到的钢球表面缺陷图像进行图像处理及特征提取,得到学习样本和预测样本,运用MATLAB软件对学习样本分析并确定合理的神经网络结构,从而精确的识别出预测样本中钢球表面缺陷的类型,通过大量实验分析,验证了该识别方法的准确性及可行性。最后在上述研究基础上,确定了光源系统、展开系统、基于单片机控制的控制系统等检测仪关键部分的设计方案,搭建了可以实际应用的基于图像技术的钢球检测仪样机,通过实验验证了所获得研究结论的正确性。更多还原

【Abstract】 Bearing is basic mechanical components, and steel balls as the key parts of rolling ball bearings while its surface defects directly affect the precision bearings, dynamic performance and service life. Therefore, there is a very important theoretical and practical significance to research surface defect detection. This article has developed source selection of detector, dynamic simulation of the expand sector, steel ball surface defect pattern recognition and others of key technical issues in-depth study, and it has built a prototype for practical applications. The main contents are as follows:Optimization studies of source in detection system have been done. Starting from analyzing reflective properties of steel ball surface, it has built the reflective model of steel ball surface. Through a large number of theoretical and experimental analyses on large area flat diffuse light sources, flat circular diffuse light source, sphere diffuse light source and coaxial light source, the combination of FPR source and LDR source has been decided as lighting schemes for testing institution. The lighting program effectively solves the problems of halo phenomena, surrounding scenery greet and inverted image of camera and so on.The results greatly improved steel ball image quality and effective detection area, which lay good foundation for the later period image processing. The deployment mechanism model of detection system has been established by combing UG and ADAMS, and kinematics and dynamics simulation are also done, simulating the actual trajectory、force and impact conditions of steel ball in expand plate detect cavity. The ball has springback movement due to collision with sidewall, the motion state of steel ball has changed because of optimization diameter of expand cavity、damping、speed of friction cavity and expand cavity、rubbing speed of friction cavity and other structural and movement parameters, thus the device ensure the steel surface can completely expand and ensure detection efficiency is highest.The key technology of defect recognition and its classification is done. First image preprocessing method of enhancing image acquisition and noise reduction of steel ball surface are researched. The original images go through two times the handling to eliminate high frequency white noise. Second the image smooth the image after denoising, and then setting gray threshold using canny operator to images on edge of the inspection. Finally, morphological image processing and image enhancement processing are done, which lay a foundation for defect feature extraction of steel ball. Defect area、defect length-diameter ratio、defect circumference and Euler number are determined as characteristic parameter, and it has proposed a method of the steel ball surface defect based on BP neural network. Collected by the steel ball surface defect images for image processing and feature extraction, by studying samples and prediction samples, to learn to analysis sample using matlab software, and determine a reasonable neural network structure, to forecast accurately identify samples of the type of steel ball surface defect, through a lot of the experimental analysis, verify the accuracy of identifying ways and feasibility.Finally, combining the problem of research results, the study determined the design of the key parts of detector such as the light sources of the system、the expand system and the control system based on SCM, and ball detector prototype based on image technology has been built, the correctness of research conclusions is verified through experiments.更多还原